Module resilience 

Resilience layer (RFC-029) — failure classification and recovery coordination.

Layers

• [classify] — maps a provider error to a FailureClass.
• budget — AttemptBudget, bounds total retries.
• coordinator — RecoveryCoordinator, the OTP-style recovery ladder (L1 backoff → L2 model swap → terminal).

Honest limitation

The typed oxi_ai::ProviderError is stringified at the oxi-agent boundary. Downcasting across that boundary is not possible today, so classify uses Display-string heuristics. See [classify] for the pattern list and its caveats.

Re-exports

pub use budget::AttemptBudget;

pub use classify::classify;

pub use coordinator::RecoveryCoordinator;

pub use coordinator::ResilienceConfig;

pub use error::AgentRunError;

pub use health::BreakerConfig;

pub use health::ProviderHealthRegistry;

Modules

budget

Attempt budget — bounds total directive executions across the recovery ladder (RFC-029 §3.4, D7).

classify

Failure classification heuristics (RFC-029 §3.3).

coordinator

Recovery coordinator (RFC-029 §3.4) — drives the OTP-style recovery policy on top of the Unix lifecycle.

error

Agent run error — carries the provider error AND the agent’s exported conversation state so the recovery coordinator can snapshot→restore rather than re-fork from scratch (RFC-029 §3.2, P2b).

health

Per-provider circuit breaker — ProviderHealthRegistry (RFC-029 §3.6).

Enums

FailureClass

What kind of failure occurred, and the recovery it implies.